1 Introduction

As the coronavirus spread in Philadelphia this year, travel patterns changed. With this shock to human mobility came another to business activity, as far fewer of us traveled to work, amenities or other pastimes. In order to understand the consequences of these changes, we use GPS data from mobile phones to explore how Philadelphians changed their travel patterns before and during the pandemic.

With the goal of understanding the time-space patterns of resident movement in Philadelphia, the following section presents data from SafeGraph, a provider of mobility data collected from iPhone and Android smartphones. Note that SafeGraph collects data on a representative sample (10%) of the population across the country, so our indicators are not the true number of visits or journeys, but a slice. The data contain the terms defined in Figure 1.1: the number of visitors is the count of devices flowing to a point of interest (POI) while a connection is an origin-destination line between a Census Block Group and a point of interest, regardless of its weight. A flow is also a connection, but with a weight measuring the number of visitors traveling between origin and destination.

1.1 Key definitions

Each visit is a mobile device entering into a point of interest; these include parks and museums, restaurants and bars, or offices and hospitals. In Figure 1.2 we map the distribution of these venues and businesses for context. We classify each point of interest by its description, which SafeGraph provides. 1 We can see that most businesses cluster in Center City or nearby but no businesses cluster more than restaurants and bars.

1.2 Distribution of points of interest

To demonstrate how this information combines to form a network, Figure 1.3 connects origins to destinations by month in Philadelphia. This network is what we will be exploring below.

1.3 Connections over time

This analysis comprises different spatial scales, Citywide, Neighborhood and Point of Interest. We can look globally, across the city, to explore trends throughout; we can also think locally, dividing the city up into cells or neighborhoods to probe variations within the city. Finally, we can look at individual businesses or venues. Below, we attempt to understand patterns at each scale. We explore each of these levels for our critical variables—first at all scales for visits and then focusing on points of interest connections.

2 How have visitations changes?

In this section we explore trends and relationships in visits manifest most strongly citywide and in neighborhoods. To see how the business environment is for chains across the city, rather than any given location, we sum visits by brand. Figure 2.1 ranks each brand by the number of visitors it received and animates this change through the pandemic. Dollar stores rise gradually throughout the year; another important shift is away from non-essential retail towards essential businesses like pharmacies. Starbucks and Wawa occupy top spots for the first several weeks of the year but when the shelter-in-place order sets in, visits collapse and they are replaced in the ranks by essentials RiteAid and ShopRite.

2.1 Comparing foot traffic by brand

In Figure 2.2 we aggregate by use, grouping by classes like leisure (restaurants and bars) and tourism (museums and theaters). The pandemic has curbed visits to each class of business, but particularly leisure and other, which includes office spaces. Interestingly, tourism is recovering while shopping and grocers are not, perhaps as many switch to digital commerce.

2.4 Rolling averages

A strength of these data is that, because visits are logged at the point of interest, we can check to see if dynamics are evident at smaller scales. Figure 2.5 shows visits by 500 by 500m grid cells. We are still aggregating from points of interest, so this is visits to businesses, parks, museums and the like, but by tile; this does not include visits to the particular patch of land without setting foot in a point of interest. The visits out during the worst months of the pandemic but the Old City, Center City, University City axis still appears to have pockets of activity in these maps, indicating that the relative positions of many cells did not change.

2.5 Visits by month in gridded units

Businesses are not evenly distributed across the city, however, so understanding business activity requires a unit of analysis that respects commercial corridors—zones where businesses cluster together—of which the city has designated roughly 280. We look at restaurants and bars as an indicator of night life in Figure 2.6. The largest are Market West and Market East, on either side of city hall, with 1712 and 1263 restaurants and bars respectively, following by Old City at 654 and another in University City with 493: most of the business activity is concentrated in a few locales.

2.6 Commercial corridors

Plotting trends in these corridors over time, the greatest reduction in visitors is in Center City and at the Sports Complex. Plazas like Oxford and Levick, home to a supermarket, and City and Haverford see the smallest impact.

3 How have connections changes?

This section looks at connections by looking at points of interest and brands, how they perform over time and whether or not we can identify certain bellwether businesses. These are specific cases that can provide further insight into how the pandemic is changing mobility. Breaking out the above animation in Figure 3.1, we can see that the drop in mobility primarily impacts businesses at the center while clusters along the periphery maintain connections.

3.1 Aggregate mobility

As we saw above, the data above show that big box stores like Target and Walmart appear to have weathered the pandemic well, but the shift to remote work should also appear in the data. We can look at visits to the Comcast Center and the Plaza below it; visits in April and May, as the coronavirus took hold in the city, fell substantially.

3.2 Comcast Center in focus

As a signal for tourism, we can look at Reading Terminal Market; vendors between on its premsies saw marked declines in visits beginning in April.

3.3 Reading Terminal in focus

These destinations are critical to Philadelphia’s office and tourist economies, but what about the brands that serve thousands on a daily basis. Large flows between a neighborhood and point of interest can drive visits but counting connections allows us to see how geographically diverse a clientele is. Figure 3.4 shows the number of connections certain brands lost between January and August selecting the top and bottom 10. It shows that brands deemed essential businesses saw comparably less of a decline than others, along with fast food restaurants. The map shows the locations of brands for context.

3.4 Relative brand performance

Rankings Locations

We can see the differential effect of the pandemic simply by looking at the change in visits to Target stores and Planet Fitness gyms. These are interesting foils because both see similar numbers of visitors prior to the arrival of coronavirus and they have a similar number of locations—12 Targets and 14 Planet Fitnesses. Connections to Target held steady throughout the pandemic, even—as we see above—many connections across the city broke.

3.5 Connections to target stores

We can compare this network and its progression through the pandemic to Planet Fitness. This brand’s gyms saw fewer connections across the city as case rates grew; importantly, this brand may be an important lens into the recovery, as we see connections grow in July and August.

3.6 Connections to planet fitness gyms

Consequences and Conclusion

Changing mobility may be exacerbating issues relating to equity. Philadelphia still shows patterns of concentrated poverty, segregated housing and isolated pockets of prosperty; the pandemic could produce deeper disparities. One risk is that communities of color and low income neighborhoods will not be able to socially distance in the same capacity as affluent communities. The evidence for this is mixed. The charts below plot change in outflows from a Census Tracts as a function of both income and race, respectively. The vertical axis shows the change in travel from a January/February baseline, so a greater reduction indicates stronger adherance to social distancing rules. During the critical month of April few poor communities could afford to shelter in place at the rate that wealthy ones could.

2.4 Did trips change with income?

In March, at the onset of the pandemic, more Philadelphians were leaving their home tracts. In April, all communties see reductions in trips. However, the higher the median income, the greater the reduction in trips. The reverse is true for race: the higher the minority population, the greater the reduction of trips, but the relationship diminishes during the summer.

2.5 Did trips change with race?

In conclusion, we find that mobility dropped citywide and across industries and brands. Even the best performing brands saw a drop in visits and the same holds true for best performing neighborhoods. Yet the mobility shock was not felt evenly: essential businesses saw smaller reductions in visitors than those in leisure and tourism. Further, while businesses in Center City and Old City saw larger reductions, those located in shopping centers fared better. Changing mobility has important consequences on equity in Philadelphia: communities with low median income send visitors to other neighborhoods at comparably high rates. The evidence we have assembled, however, suggests that this divide in capacity to socially distance is a function of income more so than race

  1. If that description contains “restaurant” or “bar”, we classify that as leisure. Anything educational, from tutoring to public, private or charter schools to tertiary education, we call that education. Tourism includes museums and parks.↩